MCP Troubleshooting

Find the message or symptom you see, then follow the fix. The messages are quoted the way your assistant receives them, so you can also search this page for the exact words.

Quick referenceCopy Anchor Link

Connection problemsCopy Anchor Link

Symptoms: adding the connector fails, the server shows red or "error" in your client, a "404" or "not found", or the client says it is not an MCP server.

Fixes:

1. Check the URL. It must be exactly https://mcp.3daistudio.com/mcp.
   • Not https://www.3daistudio.com (that is the website, not the connector).
   • Not https://mcp.3daistudio.com without /mcp at the end.
   • Not an address ending in /sse from an older guide.
2. Check the connection type. It must be a remote server using Streamable HTTP (often just called "http" or "remote"), with OAuth sign in. Remove any API key, token or custom header you added.
3. Update your client. Remote servers with sign in need a recent version of Claude Code, Codex, Cursor, VS Code or OpenCode.
4. Restart the client after changing its MCP configuration.
5. Ask your assistant. AI apps change how they add MCP servers often, so our guides can lag behind. Ask your assistant: "How do I add a remote MCP server in this app right now? Use https://docs.3daistudio.com/mcp-setup.md." Coding agents can usually fix the setup themselves.

Error on the sign in page ("Connection error"): close the window and start the connection again from your assistant. If it keeps happening, remove the connector from your assistant, add it again, and make sure the URL is correct.

Plan requiredCopy Anchor Link

Messages:

"A paid plan is required. Connecting AI assistants like Claude, ChatGPT, or Cursor to your 3D AI Studio account is available on paid plans."

"The 3D AI Studio MCP connector requires an active paid subscription. Upgrade at https://www.3daistudio.com/pricing to keep using it."

Fix: the connector is part of the paid plans. Upgrade your plan, then connect again (or just retry, if the assistant was already connected). Your existing models are never affected.

If you are part of a team, check that you chose the right credit source when connecting. See Credits & Limits.

Signed out or needs authenticationCopy Anchor Link

Symptoms: the assistant asks you to connect again, the server shows "needs authentication" or "login required", or Settings shows Sign in again.

Fix by assistant:

Why it happens: you disconnected the assistant, used Sign out of all devices, or the assistant was not used for a long time. Signing in again restores everything.

Keeps happening with Codex or another terminal tool? Close all running sessions of the tool, update it to the latest version, then sign in once more. Several old sessions running at the same time are the most common cause.

Restart neededCopy Anchor Link

Symptom: Settings shows Restart needed next to an assistant.

What it means: one window or session of that assistant is still using an old sign in. The connection itself is fine, and other sessions keep working.

Fix: close all running sessions or windows of the assistant and start a new one. For chat apps, restart the app.

Rate limitsCopy Anchor Link

Message:

"You've reached the limit of ... generations per minute for this model. Please wait ... seconds before starting another generation."

What it means: many generations with the same model were started in a short time. This limit is the same as on the website.

Fix: usually nothing. Your assistant waits the number of seconds in the message and continues. If it stops instead, say "please continue" or "try again in a minute". For big batches, ask it to generate the items one after another.

Upload limit:

"Upload limit reached (... per hour on the current plan). Please try again later."

Wait a while before uploading more images. Higher plans have a larger hourly allowance.

Not enough creditsCopy Anchor Link

Messages:

"Not enough credits for this ... generation. Current balance: ..."

"This would cost ... credits but the balance is only ..."

Fix: nothing was charged. Top up or upgrade on the pricing page, or choose a cheaper option (for example Prism 3.1 instead of P2, or a lower image quality). If you connected with team credits, the team's balance is what counts.

Image problemsCopy Anchor Link

Failed generationsCopy Anchor Link

Messages:

"... generation failed: ... Credits were refunded automatically."

"An identical recent request already failed: ... Credits were refunded. Vary the arguments (e.g. the prompt or a seed) to retry."

Fix: your credits are already back. Try again with a slightly different prompt or image, or ask for a different model (for example "try it with Hunyuan 3.1"). If the image itself is the problem, see the image tips in Images & Uploads.

"An identical generation is already in progress": the same request was sent twice. Nothing extra is charged. Wait for the first one to finish; your assistant can check it with "how is my generation doing?".

The assistant does not use 3D AI StudioCopy Anchor Link

Symptoms: the assistant writes code for a 3D shape, describes a model in words, or says it cannot create 3D models.

Fixes:

1. Enable the connector in the chat. In Claude and ChatGPT, turn on 3D AI Studio in the tools menu of the message box. In Cursor and VS Code, use agent mode and make sure the 3daistudio tools are enabled.
2. Ask explicitly: "Use 3D AI Studio to make a 3D model of ...".
3. Check the connection: ask "what's my 3D AI Studio credit balance?". If that fails, go through Connection problems and Signed out.
4. Start a new chat or session after adding the connector. Many clients only load tools at the start.

No 3D viewer in the chatCopy Anchor Link

The interactive 3D viewer appears in assistants that support interactive cards, such as Claude. Terminal and editor clients (Claude Code, Codex, Cursor, VS Code, OpenCode) show a link instead, and some chat apps show a simple card or link. In every case the model is saved in your dashboard, where you can view it in 3D and download it in any format.

Results are missing from the dashboardCopy Anchor Link

Every generation the assistant starts is saved to the dashboard of the account (and team, if you chose one) it is connected to.

• Check the team. If you connected with team credits, the results are in that team's library.
• Check the account. Make sure you signed in with the same 3D AI Studio account on the website as during the connection.
• Wait for it to finish. 3D models take 1 to 5 minutes. They appear in the dashboard right away and complete there.

Still stuck?Copy Anchor Link

Email support@3daistudio.com with:

• the assistant you use and its version (for example "Codex CLI 0.145" or "Claude desktop"),
• what you asked for and the exact message you saw,
• roughly when it happened.

Never send your password, tokens or the contents of configuration files with credentials. We never need them. For general (non MCP) problems, see the main Troubleshooting Guide.